I dont this is top most priority. We do this with codex or claude in chrome and validate the UX.
I think BDD is going to be back in style for agent generate code.
When you generate at scale discovering, maintaining and scaling test is the major problem, thats why were katana [1] a behavior driven testinf utility thay discovers behavior and maintain, generates test cases.
"It reads the screen, not the DOM", but is built completely on Playwright? At first it made me think you have some visual model at play, but doesn't actually seem that way.
Argus adds one more layer onto Playwright, so we expect it to be slower. Think of it as a kind of tradeoff where you gain human-level QA by accepting slower runs.
I think BDD is going to be back in style for agent generate code.
When you generate at scale discovering, maintaining and scaling test is the major problem, thats why were katana [1] a behavior driven testinf utility thay discovers behavior and maintain, generates test cases.
[1] https://github.com/adaptive-scale/katana
You always need real QA tests to ensure things are working smoothly.